On November 4, as the 2026 World Internet Conference Wuzhen Summit entered its second half, 20 organizations from China's agent ecosystem signed a self-discipline document at the same forum. The Agent Development Self-Discipline Pact came out of the forum on agent trust, interconnection and interoperability, and People's Daily and Securities Times both reported on it. The drafting was coordinated by the China Federation of Internet Societies, with the China Academy of Information and Communications Technology (CAICT), Tencent, Ant Group, Alibaba, Douyin and others doing the writing. As a leading domestic large-model provider with full-spectrum coverage, NineZenith Technology fully endorses the pact.
The pact implements the industry self-discipline mandate of the Implementation Opinions on the Standardized Application and Innovative Development of Agents. Securities Times summarized its framework as eight rules, covering agent trust systems and interconnection infrastructure, then higher interoperability, stronger security protection and governance mechanisms, along with user-rights protection, industry empowerment and inclusive sharing of the technology.
Where the pact carries weight is its first chapter. People's Daily reported that it calls for layered, verifiable agent identity credentials, along with mechanisms for identity verification and management, permission granting and control, and rule expression and embedding. Most agent risks reduce to identity and permission problems. Verifiable identity infrastructure is what turns agents from demos into digital employees that can be trusted with real tasks.
There are no penalties in the pact. Its binding force comes from the signatories' pledge to lead implementation and accept supervision from industry bodies and the public. A self-discipline document works by fixing a scattered consensus into a shared text, leaving hooks for future standards and regulation. The test is whether signatures hold. Outside the Wuzhen venue, whether each vendor's permission model and identity credentials actually open up to interoperability is the hard indicator to watch.
One more detail is telling. The publicly disclosed signatory list is dominated by platform, device, security and telecom players — foundation-model makers such as DeepSeek, Zhipu (GLM) and Moonshot AI are nowhere on it. How a self-discipline pact without the base-model vendors reaches the bottom of the agent stack is the question this document leaves open.
From eight rules to verifiable identity infrastructure, two hurdles remain — engineering and standards. The pact reads like the start of a roadmap. Whether the intelligent internet gets built depends on how many promises turn into interfaces in the coming six months. NineZenith will likewise align the identity and permission mechanisms of its agent products toward verifiable, interoperable practice.